Mapping and Geographical Risk Factors for Cancer in Iowa According to the Iowa Cancer Registry 2024 Cancer in Iowa Report, Iowa continues to have the 2nd highest and fastest growing rate of new cancers in the U.S. The Institute for Public Health Practice, Research and Policy has been working with the Iowa Cancer Registry to produce age-adjusted cancer rate estimates for every county in Iowa. We are interested in identifying potential risk factors related to these increases in cancer rates. New estimates of risk are available through PLACES produced by the CDC. In this project, we will relate cancer rates to various risk factors to identify potential risk factors of cancer incidence in Iowa and how those risk factors might change geographically.
Deep Learning to Predict Enrollment at the University of Iowa The University of Iowa, like many large institutions, has substantial data infrastructure regarding prospective, current, and former students. This resource is actively used by researchers in the Department of Biostatistics to help the university with recruitment, planning, resource allocation, and promoting student success and retention. This project will seek to expand the methods used to predict enrollment at the university by implementing recurrent neural network (RNN) models, and applying them to time-series of real student-level data. We will explore neural networks generally, as well as the use of graphics processing unit (GPU) accelerators and high-performance computing (HPC) resources to improve computation time.
Predicting antibiotics usage for improved healthcare-associated infection risk assessment Healthcare-associated infections (HAIs) impact an estimated 1 in 31 hospital patients and 1 in 43 nursing home residents. Many HAIs are caused by pathogens resistant to antibiotic treatments, while other HAIs arise from antibiotics suppressing probiotics (helpful microorganisms). An example is Clostridioidis difficile which in 2017 led to 223,900 infections and 12,800 deaths in the U.S. It is imperative to develop a better understanding of how these pathogens are transmitted within healthcare facilities. To help answer this question, our research group has published research on which antibiotics lead to a higher risk of a C. difficile infection (CDI), as well as how other patients with a CDI in the same facility increases one’s own risk. These two research lines come from the following two data sources respectively: (D1) over 26 million hospital encounters in which antibiotics usage is available, and (D2) over 16.8 million hospital encounters in which each patient’s healthcare facility is available. To accurately understand how facility-level features and concurrent patients with CDIs (found only in D2) combine to foster varying risk environments to newly admitted patients, we also must know the antibiotic usage of patients (found only in D1). This project, therefore, aims to use machine learning techniques to build predictive models for antibiotics found to lead to high CDI risk using D1; these will then be used to predict antibiotic usage of those patients in D2 in order to better understand how C. difficile spreads within hospitals.
Multiple Structural Breaks Detection through Genetic Algorithm Under stimuli and workloads, the human body tends to display discomforts and covert proximal cognitions that can manifest through physiological responses. One such response can be in a form of skin sweats, easily capturable via wearable sensors. The sensors capturing electrodermal activities (EDA) record big data serially (4Hz, 8Hz), per subject, over a duration of an experiment (approximately 35 min). This project aims to study serial EDA data collected on subjects from an experiment being conducted in neuropsychology at the University of Iowa, with the goal to detect structural breaks corresponding to epochs of learning activities and assess the role that biofeedback plays in efforts to engage people in a learner space. The project mobilizes technological innovations in neuroimaging (fNIRS), wearable sensors monitoring covert cognitive activity, monitoring arousal states under workload, and video data outputs to address learners’ emotional discomfort. The project will focus on structural breaks detection though a genetic algorithm stochastic search across the spectrum of EDA data.